This paper proposes a method to estimate state of charge (SoC) for Lithiumion battery pack (LIB) with π series-connected cells. The cellβs model is represented by a second-order equivalent circuit model taking into account the measurement disturbances and the current sensor bias. By using two sigma point Kalman filters (SPKF), the SoC of cells in the pack is calculated by the sum of the packβs average SoC estimated by the first SPKF and SoC differences estimated by the second SPKF. The advantage of this method is the SoC estimation algorithm performed only two times instead of π times in each sampling time interval, so the computational burden is reduced. The test of the proposed SoC estimation algorithm for 7 samsung ICR18650 Lithium-ion battery cells connected in series is implemented in the continuous charge and discharge scenario in one hour time. The estimated SoCs of the cells in the pack are quite accurate, the 3-sigma criterion of estimated SoC error distributions is 0.5%.
State of charge estimation for lithium-ion batteries connected in series using two sigma point Kalman filters
1. International Journal of Electrical and Computer Engineering (IJECE)
Vol. 12, No. 2, April 2022, pp. 1334~1349
ISSN: 2088-8708, DOI: 10.11591/ijece.v12i2.pp1334-1349 ο² 1334
Journal homepage: http://ijece.iaescore.com
State of charge estimation for lithium-ion batteries connected in
series using two sigma point Kalman filters
Chi Nguyen Van1
, Thuy Nguyen Vinh2
1
Institute of High-Technology Research and Development for Industry (RIAT), Thai Nguyen University of Technology, Thai Nguyen,
Vietnam
2
Department of Automation, Thai Nguyen University of Technology (TNUT), Thai Nguyen, Vietnam
Article Info ABSTRACT
Article history:
Received Apr 19, 2021
Revised Aug 3, 2021
Accepted Sep 1, 2021
This paper proposes a method to estimate state of charge (SoC) for Lithium-
ion battery pack (LIB) with π series-connected cells. The cellβs model is
represented by a second-order equivalent circuit model taking into account
the measurement disturbances and the current sensor bias. By using two
sigma point Kalman filters (SPKF), the SoC of cells in the pack is calculated
by the sum of the packβs average SoC estimated by the first SPKF and SoC
differences estimated by the second SPKF. The advantage of this method is
the SoC estimation algorithm performed only two times instead of π times
in each sampling time interval, so the computational burden is reduced. The
test of the proposed SoC estimation algorithm for 7 samsung ICR18650
Lithium-ion battery cells connected in series is implemented in the
continuous charge and discharge scenario in one hour time. The estimated
SoCs of the cells in the pack are quite accurate, the 3-sigma criterion of
estimated SoC error distributions is 0.5%.
Keywords:
Battery pack
Lithium-ion battery
Sigma point Kalman filter
SoC estimation
State of charge
This is an open access article under the CC BY-SA license.
Corresponding Author:
Chi Nguyen Van
Institute of High-Technology Research and Development for Industry (RIAT), Thai Nguyen University of
Technology National
666, 3/2 Street, Tich Luong Ward, Thai Nguyen City 251750, Vietnam
Email: ngchi@tnut.edu.vn
1. INTRODUCTION
Lithium-ion batteries pack (LIB) are so far known to be the most commonly used batteries with the
following advantages: a higher energy density, voltage capacity; lower self-discharge rate than other
rechargeable batteries; the price is much cheaper than the LiPo battery. LIBs power almost every portable
devices such as smartphone, laptop, and tablet sold today across the world. Now, LIBs are applied to not only
electric cars, electric motorcycles, buses, trucks but also large-scale electrical energy storage. LIB is formed
by many cells connected in series and parallel in order to get the desired voltage and power [1] as shown in
Figure 1. The largest market for LIBs has traditionally been portable electronic devices, but in the near future
there is an extensive growth in the demand for LIBs in transportation and electrical energy storages [2]. The
state of charge (SoC) is an important input for balancing, energy and power calculations of LIB. SoC
depends on the open-circuit-voltage (OCV) with the nonlinear relationship affected by the temperature,
aging, and charge/discharge mode, [3]β[5]. The OCV depends on temperature and electrode particle surface
concentration, but SoC depend on particle average concentration. In the practice, optimal safe operation
range of SoC for cells is above 20% and below 90%. Figure 2 describes a typical relationship between SoC
and OCV at -5 0
C and 25 0
C of temperature for a Lithium samsung cell. SoC is a parameter that is not
measurable directly, so it is estimated by using the current, voltage, and temperature of the cell. SoC is
needed to be estimated accurately in order to use for the SoC balance control problem. Controlling the
2. Int J Elec & Comp Eng ISSN: 2088-8708 ο²
State of charge estimation for lithium-ion batteries connected in series using β¦ (Chi Nguyen Van)
1335
operation of the battery system in order to prolong the life of the battery cells, to operate safely is a topic that
scientists are currently interested in applying LIBs in transportation and electrical energy storage. To estimate
SoC for one cell, we need the model describing the dynamic behavior of the cell and the SoC estimation
algorithm based on that model with real-time current, voltage, and temperature inputs. The accuracy of the
estimated SoC depends on the model of the cell and the SoC estimation algorithms used [6].
The model of the cell: if the complex model is used (such as an electrochemical model), the
estimated SoCs will be more accurate, but the computational burden and hardware costs will increase a lot
when the number of cells in series π is large. If the equivalent-circuit model is used, the order of that model
plays an important role. The computational complexity of the SoC estimation algorithm is simpler with the
lower-order model but the cell dynamic behavior is not reflected accurately. There is higher accuracy in SoC
estimation when the higher-order model is used, but the computational burden is increased [7]. The SoC
estimation algorithm: Up to now, most of the SoC estimation algorithms are based on coulomb counting [8],
Kalman filter (KF) [9], adaptive extended Kalman filter (EKF) [10], [11], and other [12], [13] improved
algorithms. Several other algorithms such as the use of fuzzy models and neural networks are also applied
[14], [15].
Figure 1. Inside the tesla model S battery pack
(a) (b)
Figure 2. The typical relationship between SoC and OCV for a lithium samsung cell: (a) tested at -5 0
C and
(b) tested at 25 0
C
The SoC estimation for π cell connected in series is the challenge problem, there are some works
related to this in literature [16], [17]. Normally, the SoC estimation algorithm is implemented repeatedly π
times, if the number of cells π is large then this is really a computational burden for the battery management
system [18]. Beside that, from the works [7], [12], [14], [17] and the actual testing we found that the
interference of the measuring noises, current sensor bias are an obstacle to estimating SoC exactly. In
practice, the current through the cell is usually measured by a shunt resistor with a voltage signal of ππ, this
voltage signal is more susceptible to interference than the voltage measuring signal at both ends of the cell
(2.7 π to 4.2 π). The current sensor bias of the rectifier affected by temperature needs to be considered in the
3. ο² ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 2, April 2022: 1334-1349
1336
current measurement. The KF theory assumes that all noises are zero mean, so an unknown current sensor
bias can introduce permanent SoC error. This paper introduces the SoC estimation method for a LiB
connected in series using sigma point Kalman filters (SPKF) to minimize the influence of current and
vfoltage measurement noise on the estimated SoC results. By using the second-order equivalent circuit model
(SECM), this allows to describe more exactly the cellβs dynamic in the charged and discharged scenarios
with amplitude varying suddenly (an important feature of electrical vehicles using LIBs must be considered)
[19]. This method uses two SPKFs, the SoC of cells in the pack is calculated by sum of packβs average SoC
estimated by first SPKF and SoC differences estimated by second SPKF.
The contributions of this work are: The first, by using only two filters to estimate the π SoCs of π
cells connected in series, the SoC estimation algorithm is performed only two times instead of π times in
each sampling time interval, so the computational burden reduced to π β 2 times, this is a very effective
solution when the battery pack has a large number of cells. The second, the SoC estimation taking into
account the current sensor bias in the model in association with applying the SPKF filter allows mapping the
distributions of current, voltage, and temperature signals to the distribution of the estimated SoCs of the cells
with more accuracy tested by the 3ο³ criterion. The remaining part of this paper is structured as follows:
section 2 mentions the SECM of the cell taking into account the measurement noises and the current sensor
bias. Section 3 introduces the proposed SoC estimation algorithm of cells connected in series using 02
SPKFs. Section 4 shows the results applied on the Lithium ICR18650-22P SAMSUNG cells and discussions.
Finally, section 5 summarizes some conclusions.
2. THE SECOND-ORDER EQUIVALENT CIRCUIT MODEL OF CELL
Suppose that the LIBβs cell is depicted as in Figure 3(a), in which π(π‘) is the cell current (π(π‘) > 0
for charge and π(π‘) < 0 for discharge, π(π‘) = 0 for cell in no charge/discharge), 10% < π§(π‘) < 95% is SoC,
3.6π < π£(π‘) < 4.2π is the cell voltage, β50
πΆ < π(π‘) < 500
πΆ is cell temperature. In order to estimate the
SoC at the sample time π, which is π§π, we need to measure the voltage, current and cell temperature. Let
ππ + ππ + ππ,π is the measured current, π£π + ππ£,π is measured voltage, in which ππ is the current sensor bias;
ππ,π is the current noise; ππ£,π is the voltage noise. The Figures 3(b), 3(c) are the recorded voltage and current
in the practice, respectively. The scanning electrochemical microscopy (SECM) of the cell is plotted in
Figure 4 with ππΆπ(π§(π‘), π) is the relationship between the oral cholera vaccine (OCV) and SoC, this
relationship is nonlinear function depending on cell temperature and this function is needed to determine by
the practice data collected by tests. The parameter β(π‘) is the dynamic hysteresis voltage as a function of SoC
and time, π 1(π), πΆ1(π), π 2(π), πΆ2(π) are two resistor-capacitor pairs describe the cell dynamic, π 0(π) is
series resistance parameter, π0(π), π(π) are the hysteresis parameters, π(π) is Coulombic efficiency, πΎ(π)
is the parameter related to the the rate of decay. Note that all parameters of cell depend on the cell
temperature. In exact terms, they also depend on the cell aging, the charge/discharge condition. For different
kind of cell (Samsung, Panasonic), they are different. In this work, we suppose that all parameters of cell
only depend on the cell temperature.
Suppose that all the quantities is discreterized at the sampling time π, π = 0,1,2, β¦ , β with the
sampling interval βπ‘ is small enough so that the current and voltage of cell are constant over that interval.
Suppose that π is total capacity of a cell, the change in SoC over sampling interval is (1).
π§π+1 = π§π β ππ
ππΞπ‘
π
(1)
The diffused resistor currents are computed as (2),
ππ 1,π+1 = ππ₯π (β
βπ‘
π 1πΆ1
) ππ 1,π + (1 β ππ₯π (β
βπ‘
π 1πΆ1
)) ππ
ππ 2,π+1 = ππ₯π (β
βπ‘
π 2πΆ2
) ππ 2,π + (1 β ππ₯π (β
βπ‘
π 2πΆ2
)) ππ (2)
Hysteresis voltage is given by (3).
βπ+1 = ππ₯π (β |
ππππβπ‘
π
|)βπ β (1 β ππ₯π (|
ππππβπ‘
π
|))π πππ(ππ) (3)
4. Int J Elec & Comp Eng ISSN: 2088-8708 ο²
State of charge estimation for lithium-ion batteries connected in series using β¦ (Chi Nguyen Van)
1337
(a)
(b) (c)
Figure 3. Illustration of the cell measurements, (a) the LIB, (b) the recorded voltage (c) current of cell
Figure 4. The second-order equivalent circuit model of the cell
The current sensor bias is ππ = ππβ1 + ππβ1, in which ππβ1 is virtual noise of the current sensor
bias used only for the calculation. The output equation of the model describing the relationship of the cell
voltage, the current, SoC, the hysteresis, and the diffused resistor currents is written as (4).
π¦π = ππΆπ(π§π) + πβπ β π 1ππ 1,π β π 2ππ 2,π β π 2(ππ β ππ) (4)
The state vector related to the diffused resistor currents, the hysteresis, SoC and the current sensor bias as is
formed as the (5).
xπ = [ππ 1,π+1 ππ 2,π+1 βπ+1 π§π+1 ππ+1]π
(5)
Define the system matrices and output matrix of the model as.
Aπ πΆ = [
ππ₯π (β
βπ‘
π 1πΆ1
) 0
0 ππ₯π (β
βπ‘
π 2πΆ2
)
] , Bπ πΆ = [
1 β ππ₯π (β
βπ‘
π 1πΆ1
)
1 β ππ₯π (β
βπ‘
π 2πΆ2
)
] , Aπ»,π = ππ₯π (|
ππππβπ‘
π
|)
5. ο² ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 2, April 2022: 1334-1349
1338
A(ππ) = [
1 0 0
0 Aπ πΆ 0
0 0 Aπ»,π
] , B(ππ) = [
β
ππβπ‘
π
0
Bπ πΆ 0
0 1 β Aπ»,π
] , Cπ = [0 βπ 1 βπ 2 π], Dπ = βπ 0
Suppose the input and output vectors of the model as (6).
uπ = [
ππ + ππ
π πππ(ππ + ππ)
] , π¦π = π£π (6)
Finally, the SECM of the cell is written as (7).
{
xπ+1 = A(ππ)xπ + B(ππ)uπ
π¦π = ππΆπ( π§π, ππ) + Cxπ + Duπ
(7)
Model (7) is used to estimate the SoC for the cell with input uπ and output π¦π without measurement
noise. When measuring uπ and π¦π, the two measurement noises ππ,π and ππ£,π were mixed in the curent and
voltage as described in Figure 3(a). So that, the input vector and output vector of the model is written as (8).
uπ,π = [
ππ + ππ + ππ,π
π πππ(ππ + ππ+ππ,π)
] , π¦π,π = π£π +ππ£,π (8)
With the input and output vectors in the (8), the model (7) is rewritten as (9).
{
xπ,π+1 = A(ππ,π)xπ,π + B(ππ,π)uπ,π
π¦π,π = ππΆπ( π§π,π, ππ) + Cxπ,π + Duπ,π
(9)
In which xπ,π+1 is state vector affected by measurement noise formed as (10).
xπ,π = [ππ 1π,π+1 ππ 2π,π+1 βπ,π+1 π§π,π+1 ππ,π+1]π
(10)
Suppose the error wπ = xπ+1 xπ,π+1 to be the system noise of the model, ππ = π¦πβπ¦π,π is the
output noise of the model, we have (11).
{
xπ+1 = A(ππ,π)xπ,π + B(ππ,π)uπ,π + wπ
π¦π = ππΆπ( π§π, ππ) + Cxπ + Duπ,π + ππ
(11)
The model (11) is used to estimate the state vector xπ+1, in that vector there are the SoC π§π and the current
sensor bias ππ need to be estimated. In this work, we use LIB Samsung ICR18650-22P [20] with parameters
π = 2200 ππ΄β, π
πππ₯ = 4.25 π, ππππ = 2.75 π, nominal voltage 3.62 π. The test temperature range is
considered from β5 Β°πΆ to 45 Β°πΆ. The values of modelβs parameters determined by practical tests are given in
Table 1.
Table 1. The cellβs model parameters of LIB Samsung ICR18650-22P
T0
C -5 0
C 5 0
C 15 0
C 25 0
C 35 0
C 45 0
C
ο¨π€
1.0869 0.9803 1.0220 0.95 1.0542 1.0399
π (ππ‘) 2.1596 2.1877 2.1943 2.175 2.1515 2.1523
ο§ 243.1946 78.4915 63.6762 250.0074 170.6407 151.3064
ππ (π) 0.0072 0.0049 0.0035 0.002 0.0036 0.0024
π (π) 0.0347 0.0257 0.0218 0.02 0.0201 0.0185
ππ (β¦) 0.013 0.023 0.0292 0.033 0.0012 0.0011
ππππ (π¬) 0.6124 1.7555 0.3227 0.899 0.6997 0.4630
ππππ (π¬) 3.9035 7.5994 8.1118 10.023 8.1840 6.5319
ππ (β¦) 0.0204 0.0203 0.0201 0.019 0.019 0.0019
ππ (β¦) 0.0494 0.0376 0.0288 0.022 0.0136 0.0134
6. Int J Elec & Comp Eng ISSN: 2088-8708 ο²
State of charge estimation for lithium-ion batteries connected in series using β¦ (Chi Nguyen Van)
1339
3. PROPOSED SOC ESTIMATION ALGORITHM FOR π΅ CELLS CONNECTED IN SERIES
USING TWO SPKFS
Suppose that there is π cells LIB connected in series as shown in Figure 5, the problem is to
estimate SoCs of π cells with the input and output signals affected by noises as (8), the noise distributions
are considered as Gaussian-distributed. SPKF is the filter with high precision aimed at nonlinear system, in
practice, the SPKF has some advantages compared to the EKF. This filter allows performance KF for
nonlinear systems without the linearization steps. The set of weighted samples, called sigma points, is used
for normalizing mean and covariance of a probability distribution. The sigma points of input distributions are
mapped thought the nonlinear dynamic model functions to get the sigma points of output distribution
[21], [22]. SPKF allows estimation states of nonlinear dynamic systems more accurately than the EKF does,
but the computational complexity is same as EKF [23]β[25]. SPKF is easily implemented because there is no
requirement of the Jacobi matrix calculation. The accuracy of state estimation using the SPKF mainly
depends on the strategies of choosing sigma-points [26], [27]. Normally, we need to estimated SoC π times
for π cells. If the number of cells π is large, there is a large amount of calculation [18], [19]. To reduce the
SoC estimation times, in this work we use only two SPKFs, the first filter estimates the mean of π SoC and
the current sensor bias, the second filter estimates π SoC differences from the average SoC, the estimated
SoC of each cell then is calculated by the sum of the average SoC determined by the first filter and the
difference SoC estimated by the second filter. In one sampling time interval, the SoC estimation algorithm is
implemented two times instead of π times.
Figure 5. π LIB cells connected in series
Consider the second model of cell in the (11) with input and output in the (8). The SOC estimation
algorithm flowchart for π cell connected in series is conducted as:
The SOC estimation algorithm flowchart
Begin
Initialization
For each sampling time π, π = 1,2,3, β¦ , β, do
Measure the current of cell ππ,π ; Measure voltages of π cells π¦π,π
1
, π¦π,π
2
, β¦ , π¦π,π
π
;
Measure ambient temperature of cell ππ
Estimate state vector x
Μπ = [πΜπ 1,π+1 πΜπ 2,π+1 β
Μπ+1 π§Μπ+1 π
Μπ+1]
π
by first SPKF
For each cell π = 1,2,3, β¦ , π do
Estimate the SOC difference βπ§Μπ
(π)
by second SPKF; Estimate
the SoC for cell π§Μπ
(π)
= π§Μπ + βπ§Μπ
(π)
End
End
End
3.1. Algorithm to estimate the average SoC for the N cells by the first SPKF filter
The first filter estimates the mean of π SoC and the current sensor bias as steps following:
β Form the augmented state vector at sampling time π defined by x
Μπβ1
π+
, including the estimated state
vector at the sampling time π β 1 the mean vector of the system noise w
Μ and mesurement noise ΞΎΜ :
x
Μπβ1
π,+
= [x
Μπβ1
+
w
Μ ΞΎΜ ] (12)
β Form the augmented covariance matrix (CM) of state estimation error vector by using CM of state
estimation error at sampling time π β 1, the CM of system noises and the CM of measurement noises:
7. ο² ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 2, April 2022: 1334-1349
1340
Οx
Μ,πβ1
π,+
= ππππ[Οx
Μ,πβ1
+
Οw Οπ] (13)
β Set up matrix of π + 1 sigma points:
Xπβ1
π,+
= {x
Μπβ1
π,+
, x
Μπβ1
π,+
+ πΎβΟx
Μ,πβ1
π,+
, x
Μπβ1
π,+
β πΎβΟx
Μ,πβ1
π,+
, } β [Xπβ1
x,+
Xπβ1
π€,+
Xπβ1
ΞΎ,+
]
π
(14)
β Calculate sigma points matrix at sampling time π by using the cell modelβs dynamic function in (11):
Xπ,π
x,β
= Aπ(ππ,π)Xπβ1,π
x,+
+ Bπ(ππ,π)uπ,π+Xπβ1,π
π€,+
π = 0,1,2, β¦ π (15)
β Estimate the priori state vector at the sampling time π:
x
Μπ
β
= β πΌπ
π
Xπ,π
x,β
π
π=0 (16)
β Update the CM of the state estimation errors:
Οx
Μ,π
β
= β πΌπ
π
(Xπ,π
x,β
β x
Μπ
β
)
π
π=0 (Xπ,π
x,β
β x
Μπ
β
)
π
(17)
β Calculate the output sigma points matrix at the sampling time π by using the output equation in the
model (11):
Yπ = ππΆπ(π§Μπ, ππ) + CπXπ,π
π₯,β
+ Dπuπ,π + Xπβ1
ΞΎ,+
(18)
β Estimate the output voltage of cells as (19):
y
Μk = β πΌπ
π
π
π=0 Yπ (19)
β Update the CM of output voltage estimation errors:
Οy,
Μπ
β
= β πΌπ
π
π
π=0 (Yπ β y
Μk )(Yπ β y
Μk )π
(20)
β Update the covariance matrices of the state vector estimation error and output voltage estimation error:
Οx
Μ,π¦,
Μπ
β
= β πΌπ
π
(Xπ,π
x,β
β x
Μπ
β
)
π
π=0 (Yπ β y
Μk )π
(21)
β Update the estimation gain matrix as (22):
Lπ =
Οx
Μ,π¦,
Μ π
β
Οy,
Μπ
β (22)
β Update the estimation state vector by using the estimation gain matrix, the measured real output
voltages and the estimated output voltage of cells:
x
Μπ
+
= x
Μπ
β
+ Lπ (yπ β y
Μk ) (23)
β Update the CM of state estimation error vector:
Οx
Μ,π
+
= Οx
Μ,π
β
β LπΟy,
Μπ
β
Lπ
π
(24)
3.2. SoC differences estimation for π΅ cells by second SPKF
The second SPKF estimates the SoC differences βπ§Μπ
(π)
, π = 1,2, β¦ , π, for π cells at the sampling
time π as following steps:
8. Int J Elec & Comp Eng ISSN: 2088-8708 ο²
State of charge estimation for lithium-ion batteries connected in series using β¦ (Chi Nguyen Van)
1341
β Form the augmented state vector of cell π at the sampling time π β 1 by using the SoC difference βπ§Μπ
(π)
and the mean vector of the SoC difference noises π€
Μ π§β
(π)
:
βπ§Μπβ1
π,(π)+
= [βπ§Μπ
(π)
, π€
Μ π§β
(π)
] (25)
β From the augmented CM of SoC difference estimation error of cell π, by using the CM of SoC
difference estimation error of cell π at sampling time π β 1 and the CM of SoC difference noise:
πβπ§,
Μπβ1
π,(π)+
= ππππ (πβπ§,
Μπβ1
(π)+
, ππ€βπ§
(π) ) (26)
β Set up the matrix of π + 1 sigma points of cell π:
βππβ1
π,(π)+
= {βπ§Μπβ1
π,(π)+
, βπ§Μπβ1
π,(π)+
+ πΎβπβπ§
Μ,πβ1
π,(π)+
, βπ§Μπβ1
π,(π)+
β πΎβπβπ§
Μ,πβ1
π,(π)+
β {βππβ1,π
βπ§,(π)+
, βππβ1,π
π€βπ§
+
}
π
(27)
β Calculate the state sigma point matrix of cell π at the sampling time π:
βππ,π
βπ§,(π)β
= βππβ1,π
βπ§,(π)+
β
((ππ,πβπΜπ,π)+βππβ1,π¦
π€βπ§
+
)
3600π
π = 0,1,2, . . , π (28)
β Estimate the SoC difference of cell π at the sampling time:
βπ§Μπ
(π)β
= β πΌπ
π
βππ,π
βπ§,(π)
π
π=0 (29)
β Update the CM of SoC difference estimation error of cell:
Οβπ§
Μ,π
(π)β
= β πΌπ
π
(βππ,π
βπ§,(π)β
β βπ§Μπ
(π)β
)
π
π=0 (βππ,π
βπ§,(π)β
β βπ§Μπ
(π)β
)
π
(30)
β Calculate the output sigma points matrix of cell π:
Yπ
(π)
= ππΆπ(π§Μ π + βπ§Μπ
(π)
, ππ) β ππ,πβπ (π)
+
π(π)
β
Μπ β π 1
(π)
πΆ1
(π)
πΜπ ,1 β π 2
(π)
πΆ2
(π)
πΜπ ,2 β π 0(ππ,π β πΜπ
π
) (31)
β Estimate the output voltage of cell π:
π¦
Μπ
(π)
= β πΌπ
π
π
π=0 Yπ
(π)
(32)
β Update the CM of output voltage estimation error of cell π:
ππ¦,
Μ π
(π)
= β πΌπ
π
π
π=0 (Yπ
(π)
β π¦
Μπ
(π)
)(Yπ
(π)
β π¦
Μπ
(π)
)
π
(33)
β Update the CM of SoC difference estimation error and output voltage estimation error of cell π:
πβπ§
Μ,π¦
Μ,π
(π)β
= β πΌπ
π
(βππ,π
βπ§,(π)β
β βπ§Μπ
(π)β
)
π
π=0 (Yπ
(π)
β π¦
Μπ
(π)
)
π
(34)
β Update the state estimation matrix of cell:
πΏπ
(π)
=
πβπ§
Μ,π¦
Μ,π
(π)β
ππ¦
Μ ,π
(π) (35)
β Estimate the SoC difference of:
9. ο² ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 2, April 2022: 1334-1349
1342
βπ§Μπ
(π)+
= βπ§Μπ
(π)β
+ πΏπ
(π)
(π¦π,π
(π)
βπ¦
Μπ
(π)
) (36)
β Update the CM of SoC difference estimation error of cell π:
πβπ§
Μ,π
(π)+
= πβπ§
Μ,π
(π)β
β πΏπ
(π)
ππ¦,
Μ π
(π)
πΏπ
(π)π
(37)
3.3. SoC estimation of cells
The SoC of cell π calculated as (38):
π§Μπ
(π)+
= π§Μπ
+
+ βπ§Μπ
(π)+
(38)
4. RESULTS AND DISCUSSION
The experimental system of SoC estimation is described in Figure 6, the components in the
experimental system are listed as: i) The LiB samsung is formed by 7 cells connected in series with
parameters listed in Table 1; measurement modul is created by 7 voltage difference measurement circuits, 1
current measurement curcuit, 1 temperature sensor. The voltage difference signals are to the digital signals
10 bit and transferred to the personal computer (PC) by RS485; ii) the PC implements the SoC estimation
program for cells, the configuration of PC is core (TM) i3-6100 CPU@3.70 GHZ, 4 cores, RAM 8 GB; the
MATLAB software is used to implement the SPKF algorithms, the LabVIEW 2018 [22] is used to make the
graphic user interface, the MATLAB software is embedded to the LabVIEW to run the SoC estimation;
iii) the load is motor BLDC, YONG-d67 240 W, 24 V, 10 A, the maximum speed is 295 rmp; iv) the charge
source is owon, 195 W with 2 programmable channel 36 V, maximum current 6 A; and v) the SoC estimation
algorithms are coded in MATLAB script, then embedded to the LabVIEW environment. The sampling time
interval is 1 second. The algorithm flowchart of SoC estimation for π cell in shown in Figure 7.
Figure 6. The experimental system of SoC estimation
Figure 7. The algorithm flowchart of SoC estimation
10. Int J Elec & Comp Eng ISSN: 2088-8708 ο²
State of charge estimation for lithium-ion batteries connected in series using β¦ (Chi Nguyen Van)
1343
4.1. The SoC estimation results for 1 cell
In this part, the SoC estimation results for 1 cell is presented, the Figure 8(a) describes the current,
voltage and temperature of the cell. The noise distributions of measured currrent, voltage and temperature are
plotted in Figure 8(b), the Figure 8(c) and the Figure 8(d). The current noise was distributed inside from
β0.8 π΄ to 0.8 π΄ caused by shunt-resistor with the sensitivity 15 ππ/π΄. The range of the voltage noise is
range of -0.03 π to 0.03 π, the distribution of temperature noise is in the range from β0.7 0
πΆ to 0.7 0
πΆ.
Figure 9 shows the results obtained when performing the SoC estimation algorithm corresponding to current,
temperature and voltage inputs as shown in Figure 8 using the second-order model of cell with the
parameters described in Table 1.
The Figure 9(a) describes the comparison between real cell voltage and estimated voltage calculated
by cell model, the voltage error is plotted in Figure 9(b) with 0.1 π and β0.075 π of maximum for discharge
and charge, respectively. This maximum error is caused when there is the changing of charge to discharge
vice versa. The cause of voltage error is the dependence of model on the temperature varying in the test, the
parameters of cell were determined at particular temperatures β 5 0
πΆ, 0 0
πΆ, 5 0
πΆ, 10 0
πΆ, 15 0
πΆ, 20 0
πΆ,
25 0
πΆ, 30 0
πΆ, 35 0
πΆ. In the test, the temperature of cell raises from 25.5 0
πΆ at to 31.5 0
πΆ as plotted in
Figure 9(c). The Figure 9(c) also plotted the estimated SoC of cell. After 1 hour with the discharging (3 π΄)
and the charging (1.5 π΄) alternately, The SoC of cell tends to decrease from 92.5% at beginning to 65% at
the end of the test. The SoC estimation error is shown in Figure 9(d) by comparing to criteria π§Μπ Β± 3π.
The criteria 3π decreases gradually from the beginning of the test to the end of the test as plotted in
Figure 9(e), we can see that the accuracy of SoC estimation of the cell described by the criteria 3Ο at the end
of test was 0.0125%. The Figure 9(f) plots the updating of matrix ππ₯
Μ
+
by the time. At the times, there are the
changings of charge to discharge vice versa, the voltage errors are large, some elements of matrix ππ₯
Μ
+
need to
be changed in order to keep estimating SoC smoothly.
(a) (b)
(c) (d)
Figure 8. The measurements and distributions, (a) the current, voltage and temperature of cell in the test,
(b) the noise distributions of current, (c) temperature, and (d) voltage
Voltage (V)
Current
(A)
Temperature
(
o
C)
Voltage
(V)
Temperature
(o
C)
Current (A)
11. ο² ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 2, April 2022: 1334-1349
1344
(a) (b)
(c)
(d) (e)
(f)
Figure 9. The SoC estimation results for 1 cell, (a) the voltage and estimated voltage, (b) voltage error, (c) the
estimated SoC and temperature, (d) estimated SoCΒ±3Ο, (e) criteria 3Ο, , and (f) the matrix ππ±
Μ,π
+
+3sigma
Voltage
(V)
Current (A)
SoChat
(%)
SoChat
-3sigma
The
matrix
π
π±
Μ
,π
+
Time (s)
Voltage
(V)
SoChat
(%)
Current
(A)
Temperature (o
C)
12. Int J Elec & Comp Eng ISSN: 2088-8708 ο²
State of charge estimation for lithium-ion batteries connected in series using β¦ (Chi Nguyen Van)
1345
4.2. The SoC estimation results for 7 cells
In this part, the SoC estimation results for 7 cells connected in series are presented. Suppose that all
cells are the same with parameters listed in Table 1. The Figure 10(a) plots the cells current, the Figure 10(b)
and Figure 10(c) depicts the voltages of cells and temperature, respectively, the distributions of noises in
current, voltages and temperature are plotted in Figures 10(d), 10(e) and 10(f). The estimated SoC for 7 cells
is presented in Figure 11 and Figure 12 (in appendix), in which the Figure 11(a) is the estimated average SoC
for 7 cells, the Figure 11(b) plots the estimated SoC differences of cells.
(a) (b)
(c) (d)
(e) (f)
Figure 10. The measurements and distributions, (a) the current of 7 cells, (b) the voltages of cells, (c) the
temperature of cell, (d) the distributions of noise in current, (e) voltages and (f) temperature are plotted
The Figure 12 plots the estimated SoC of each cell and comparing with criteria 3Ο, in which:
Figure 12(a) and Figure 12(b) are the estimated SoC and comparing with criteria 3Ο for cell 1, similarity for
the cells from 2 to 7 are plotted in Figure 12(c) to Figure 12(o). The Figure 13(a) shows the varying of matrix
ππ§
Μ
+
in the test, the Figure 13(b) plots the varying of criteria 3ο³. The mean of the criteria 3ο³ for cells at the
end of the test is approximately 0.5%. Figures 12 and 13 see in appendix.
Voltage of cell 1
Voltage of cell 2
Voltage of cell 3
Voltage of cell 4
Voltage of cell 5
Voltage of cell 6
Voltage of cell 7
13. ο² ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 2, April 2022: 1334-1349
1346
(a) (b)
Figure 11. The estimated SoC for 7 cells, (a) the estimated average SoC, and (b) the estimated SoC difference
of cells
5. CONCLUSION
This paper deals with the method for SoC estimation of LIB with cells connected in series. In this
works, the cellβs model is represented by the second-order equivalent circuit model taking into account the
measurement disturbances and the current sensor bias. Using 02 SPKF filters to estimate the average SoC of
cells, the current sensor bias and the SoC differences, the estimated SoC of each cell is is calculated by sum
of packβs everage SoC estimated by first SPKF and SoC differences estimated by second SPKF. This allows
reducing the SoC estimation times to 2 times, this is a very effective solution when the battery pack has a
large number of cells. The practical test is applied to 7 samsung ICR18650-22P lithium-Ion battery cells with
the discharging (3 π΄) and the charging (1.5 π΄) alternately in 1 hour. The estimated SoC is quite accuracy
with the criteria 3 ο³ of estimated SoC error is approximately 0.5. In this work, we suppose that all cells are
physically identical, the equivalent circuit model with parameters listed in Table 1 are linearized at some
particular temperatures and applied for all cells. This is clear evidence that those factors contribute to the SoC
and current sensor bias estimation error observed in the test. In future work, the model of each cell to be
updated adaptively by temperature will be researched.
APPENDIX
(a) (b) (c)
(d) (e) (f)
Voltage 2 (V)
-3sigma
Estimated SoC of
cell 3
+3sigma
-3sigma
Estimated
SoC of cell 2
+3sigma
-3sigma
Estimated SoC of
cell 1
+3sigma
Volrage 3 (V)
Estimated SoC of cell 3 (%)
Temperature (Β°C)
Estimated
SoC of
cell 2(%)
Temperature (Β°C)
Voltage 1 (V)
Estimated SoC
of cell 1 (%)
Temperature (Β°C)
Voltage
1
(V)
Temperature (Β°C)
Estimated
SoC
of
cell
1
(%)
Estimated
SoC
of
cell
1
(%)
Estimated
SoC
of
cell
2
(%)
Temperature (Β°C)
Voltage
2
(V)
Voltage
3
(V)
Estimated
SoC
of
cell
3
(%)
Temperature (Β°C)
14. Int J Elec & Comp Eng ISSN: 2088-8708 ο²
State of charge estimation for lithium-ion batteries connected in series using β¦ (Chi Nguyen Van)
1347
(g) (h) (i)
(j) (k) (l)
(m) (n)
Figure 12. The estimated SoC for each cell and the comparing with criteria 3ο³, (a) the measurements for cell
1, (b) estimated SoC for cell 1, (c) the measurements for cell 2, (d) estimated SoC for cell 2, (e) the
measurements for cell 3, and (f) estimated SoC for cell 3, (g) the measurements for cell 4, (h) estimated SoC
for cell 4, (i) the measurements for cell 5, (j) estimated SoC for cell 5, (k) the measurements for cell 6, (l)
estimated SoC for cell 6, (m) the measurements for cell 7, and (n) estimated SoC for cell 7
(a) (b)
Figure 13. The matrix Οz
Μ
+
and criteria 3 ο³, (a) the varying of the matrix Οz
Μ
+
in the test and (b) criteria 3 ο³
Estimated SoC of cell 7 (%)
Estimated
SoC of cell 5
+3sigma
-3sigma
Estimated
SoC of cell 6
+3sigma
-3sigma
Voltage 6 (V)
Estimated SoC of cell 6 (%)
Temperature (Β°C)
Voltage 4 (V)
Temperature (Β°C)
Voltage 5 (V)
Temperature (Β°C)
Estimated
SoC of cell 4
+3sigma
-3sigma
Estimated SoC of cell 4 (%) Estimated SoC of cell 5 (%)
Voltage
4
(V)
Estimated
SoC
of
cell
4
(%)
Temperature (Β°C)
Estimated
SoC
of
cell
5
(%)
Voltage
5
(V)
Temperature (Β°C)
Temperature (Β°C)
Estimated
SoC
of
cell
6
(%)
Voltage
6
(V) Temperature (Β°C)
Voltage 7 (V)
Estimated
SoC of cell 7
+3sigma
-3sigma
Temperature (Β°C)
Estimated
SoC
of
cell
7
(%)
Estimated
SoC
of
cell
7
(%)
Voltage
7
(V)
15. ο² ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 2, April 2022: 1334-1349
1348
ACKNOWLEDGEMENTS
This research was supported by Thai Nguyen University of Technology, TNUT, Vietnam.
REFERENCES
[1] J. Warner, βLithium-ion battery packs for EVs,β in Lithium-Ion Batteries, Elsevier, 2014, pp. 127β150.
[2] K. M. Venkatachalam and V. Saravanan, βPerformance evaluation and load demand management of grid connected hybrid wind-
solar-battery system,β International Journal of Applied Power Engineering (IJAPE), vol. 9, no. 3, pp. 223β244, Dec. 2020, doi:
10.11591/ijape.v9.i3.pp223-244.
[3] X. Lin, βTheoretical analysis of battery SOC estimation errors under sensor bias and variance,β IEEE Transactions on Industrial
Electronics, vol. 65, no. 9, pp. 7138β7148, Sep. 2018, doi: 10.1109/TIE.2018.2795521.
[4] C. Yuan, Y. Deng, T. Li, and F. Yang, βManufacturing energy analysis of lithium ion battery pack for electric vehicles,β CIRP
Annals, vol. 66, no. 1, pp. 53β56, 2017, doi: 10.1016/j.cirp.2017.04.109.
[5] D. N. T. How, M. A. Hannan, M. S. Hossain Lipu, and P. J. Ker, βState of charge estimation for lithium-ion batteries using
model-based and data-driven methods: a review,β IEEE Access, vol. 7, pp. 136116β136136, 2019, doi:
10.1109/ACCESS.2019.2942213.
[6] Y. Zhou and X. Li, βOverview of lithium-ion battery SOC estimation,β in 2015 IEEE International Conference on Information
and Automation, Aug. 2015, pp. 2454β2459, doi: 10.1109/ICInfA.2015.7279698.
[7] J. Rivera-Barrera, N. MuΓ±oz-Galeano, and H. Sarmiento-Maldonado, βSoC estimation for lithium-ion batteries: review and future
challenges,β Electronics, vol. 6, no. 4, p. 102, Nov. 2017, doi: 10.3390/electronics6040102.
[8] J. Xie, J. Ma, and K. Bai, βEnhanced coulomb counting method for state-of-charge estimation of lithium-ion batteries based on
peukertβs law and coulombic efficiency,β Journal of Power Electronics, vol. 18, no. 3, pp. 910β922, 2018.
[9] P. Manoharan, M. R, K. K, and S. R, βSoC estimation and monitoring of li-ion cell using Kalman-Filter Algorithm,β Indonesian
Journal of Electrical Engineering and Informatics (IJEEI), vol. 6, no. 4, Dec. 2018, doi: 10.11591/ijeei.v6i4.548.
[10] M. Souaihia, B. Belmadani, and R. Taleb, βA robust state of charge estimation for multiple models of lead acid battery using
adaptive extended Kalman filter,β Bulletin of Electrical Engineering and Informatics (BEEI), vol. 9, no. 1, pp. 1β11, Feb. 2020,
doi: 10.11591/eei.v9i1.1486.
[11] Z. L. Huan, T. Tomohiro, and A. Tofael, βLeader-follower tracking system for agricultural vehicles: Fusion of laser and odometry
positioning using extended kalman filter,β IAES International Journal of Robotics and Automation (IJRA), vol. 4, no. 1, pp. 1β18,
Mar. 2015, doi: 10.11591/ijra.v4i1.pp1-18.
[12] V. Sangwan, R. Kumar, and A. K. Rathore, βState-of-charge estimation for li-ion battery using extended Kalman filter (EKF) and
central difference Kalman filter (CDKF),β in 2017 IEEE Industry Applications Society Annual Meeting, Oct. 2017, pp. 1β6, doi:
10.1109/IAS.2017.8101722.
[13] N. H. Mohd Amin, M. R. Ab Ghani, A. Jidin, Z. Jano, and T. Sutikno, βA review of hybrid battery management system (H-BMS)
for EV,β TELKOMNIKA (Telecommunication Computing Electronics and Control), vol. 16, no. 3, pp. 1004β1012, Jun. 2018, doi:
10.12928/telkomnika.v16i3.8743.
[14] W. Zheng, B. Xia, W. Wang, Y. Lai, M. Wang, and H. Wang, βState of charge estimation for power lithium-ion battery using a
fuzzy logic sliding mode observer,β Energies, vol. 12, no. 13, p. 2491, Jun. 2019, doi: 10.3390/en12132491.
[15] M. Charkhgard and M. Farrokhi, βState-of-charge estimation for lithium-ion batteries using neural networks and EKF,β IEEE
Transactions on Industrial Electronics, vol. 57, no. 12, pp. 4178β4187, Dec. 2010, doi: 10.1109/TIE.2010.2043035.
[16] F. Sun, R. Xiong, and H. He, βA systematic state-of-charge estimation framework for multi-cell battery pack in electric vehicles
using bias correction technique,β Applied Energy, vol. 162, pp. 1399β1409, Jan. 2016, doi: 10.1016/j.apenergy.2014.12.021.
[17] N.-T. Tran, A. Khan, T.-T. Nguyen, D.-W. Kim, and W. Choi, βSOC estimation of multiple lithium-ion battery cells in a module
using a nonlinear state observer and online parameter estimation,β Energies, vol. 11, no. 7, Jun. 2018, doi: 10.3390/en11071620.
[18] D. Saji, P. S. Babu, and K. Ilango, βSoC estimation of lithium ion battery using combined coulomb counting and fuzzy logic
method,β in 2019 4th International Conference on Recent Trends on Electronics, Information, Communication & Technology
(RTEICT), May 2019, pp. 948β952, doi: 10.1109/RTEICT46194.2019.9016956.
[19] A. M. Alsabari, M. K. Hassan, A. CS, and R. Zafira, βModeling and validation of lithium-ion battery with initial state of charge
estimation,β Indonesian Journal of Electrical Engineering and Computer Science (IJEECS), vol. 21, no. 3, pp. 1317β1331, Mar.
2021, doi: 10.11591/ijeecs.v21.i3.pp1317-1331.
[20] S. Soeprapto, R. N. Hasanah, and T. Taufik, βBattery management system on electric bike using Lithium-Ion 18650,β
International Journal of Power Electronics and Drive Systems (IJPEDS), vol. 10, no. 3, pp. 1529β1537, Sep. 2019, doi:
10.11591/ijpeds.v10.i3.pp1529-1537.
[21] W. Fan and Y. Li, βAccuracy analysis of sigma-point Kalman filters,β in 2009 Chinese Control and Decision Conference, Jun.
2009, pp. 2883β2888, doi: 10.1109/CCDC.2009.5192691.
[22] B. N. Kumar Reddy, N. Suresh, and J. V. N. Ramesh, βA gracefully degrading and energy-efο¬cient FPGA programming using
LabVIEW,β International Journal of Reconfigurable and Embedded Systems (IJRES), vol. 5, no. 3, pp. 165β175, Nov. 2016, doi:
10.11591/ijres.v5.i3.pp165-175.
[23] Qin Wang, C. Rizos, Yong Li, and Shiyi Li, βApplication of a sigma-point Kalman filter for alignment of MEMS-IMU,β in 2008
IEEE/ION Position, Location and Navigation Symposium, 2008, pp. 44β52, doi: 10.1109/PLANS.2008.4569968.
[24] C. Nguyen Van, βState estimation based on sigma point kalman filter for suspension system in presence of road excitation
influenced by velocity of the car,β Journal of Control Science and Engineering, vol. 2019, pp. 1β16, Nov. 2019, doi:
10.1155/2019/6898756.
[25] C. Nguyen Van and T. Nguyen Vinh, βSOC estimation of the lithium-ion battery pack using a sigma point kalman filter based on
a cellβs second order dynamic model,β Applied Sciences, vol. 10, no. 5, Mar. 2020, doi: 10.3390/app10051896.
[26] O. Bayasli and H. Salhi, βThe cubic root unscented kalman filter to estimate the position and orientation of mobile robot
trajectory,β International Journal of Electrical and Computer Engineering (IJECE), vol. 10, no. 5, pp. 5243β5250, Oct. 2020, doi:
10.11591/ijece.v10i5.pp5243-5250.
[27] V. D. Merwe and R. Sigma, βPoint Kalman filters for probabilistic inference in dynamic state-space models,β Oregon Health and
Science University, 2004.
16. Int J Elec & Comp Eng ISSN: 2088-8708 ο²
State of charge estimation for lithium-ion batteries connected in series using β¦ (Chi Nguyen Van)
1349
BIOGRAPHIES OF AUTHORS
Chi Nguyen Van received the PhD degree in Instrument and Control Engineering
from Hanoi University of Science and Technology (HUST) in 2012. He is currently working
as a Associate Professor with Institute of High-Technology Research and Development for
Industry (RIAT), Thai Nguyen University of Technology (TNUT), Viet Nam. His research
interests include nonlinear control, adaptive control, optimal control and system identification
for Lithium Ion Batterry. He can be contacted at email: ngchi@tnut.edu.vn
Thuy Nguyen Vinh received the MSs degree in Automation Engineering from
Thai Nguyen University of Technology (TNUT) in 2005. He is currently working as a
researcher with Institute of High-Technology Research and Development for Industry (RIAT),
Thai Nguyen University of Technology (TNUT) His research interests optimal control and
system identification for Lithium Ion Batterry. He can be contacted at email: nguyenvinhthuy-
tdh@tnut.edu.vn